public bool Validate(ITrail expected, ITrail actual, float threshold)
{
var expectedSamples = expected.GetSampledLocations();
var actualSamples = expected.GetSampledLocations();
var diff = expectedSamples.Zip(actualSamples, (e, a) => e - a);
var normDiff = diff.Select(d => d.magnitude);
var sumNormDiff = normDiff.Aggregate(0.0, (x, y) => x + y);
var mse = sumNormDiff / diff.Count();
bool isValid = mse < threshold;
return(isValid);
}
public bool Validate(ITrail expected, ITrail actual, float threshold)
{
var expectedSamples = expected.GetSampledLocations();
var actualSamples = actual.GetSampledLocations();
var diff = expectedSamples.Zip(actualSamples, (e, a) => e - a);
var normDiff = diff.Select(d => d.magnitude);
var maxNormDiff = normDiff.Aggregate(Mathf.NegativeInfinity, (x, y) => Mathf.Max(x, y));
bool isValid = maxNormDiff < threshold;
return(isValid);
}
public bool Validate(ITrail expected, ITrail actual, float threshold)
{
var expectedSamples = expected.GetSampledLocations();
var actualSamples = actual.GetSampledLocations();
var(np_exp, np_act) = prepare_np_input(expectedSamples, actualSamples);
// we know that the curves are sampled at same time intervals. If we didn't we would need to interpolate
// both of them (cubic spline or other) and sample at equal times or referably at equal distances.
// Since we do know, we can skip this phase.
// Resampling at same-spacial-distances could help, but it would be negligible here.
//normalize both curves to zero mean to be able to find curves that look the same but are far away
var one = new int[1]; one[0] = 1;
var exp_mean = np.mean(np_exp, one);
var exp_normalized = np_exp - exp_mean;
var act_mean = np.mean(np_act, one);
var act_normalized = np_act - act_mean;
// now use rigid body transform optimization, to find the matrix that registers both curves.
// this can be done because we have ordering on all points.
// http://nghiaho.com/?page_id=671
var H = np.dot(exp_normalized, act_normalized.T); // H is 2x2, if it is nxn then flip the order, and get R_act_to_exp
var(U, S, V) = np.linalg.svd(H);
if ((double)np.linalg.det(V) < 0)
{
V[":, 2"] = -V[":, 2"];
}
var R_exp_to_act = np.dot(V, U.T);
//apply rotation on exp and calculate mse
var exp_rotated = np.dot(R_exp_to_act, exp_normalized);
var error = np.linalg.norm(exp_rotated - act_normalized);
return(error < threshold);
}
public bool Validate(ITrail expected, ITrail actual, float threshold)
{
var expectedSamples = expected.GetSampledLocations();
var actualSamples = actual.GetSampledLocations();
int count = expectedSamples.Count();
for (int i = 0; i < count / 2; i++)
{
var diff = expectedSamples.Skip(i).Zip(actualSamples.Take(count - i), (e, a) => e - a);
var normDiff = diff.Select(d => d.magnitude);
var sumNormDiff = normDiff.Aggregate(0.0, (x, y) => x + y);
var mse = sumNormDiff / diff.Count();
bool isValid = mse <= threshold;
Debug.Log($"mse: {mse}");
if (isValid)
{
return(true);
}
}
return(false);
}
